Appreciate your kindness.
You're right. What I want to know is the forward pass.
Can you cross check if the calculation is correct?
Case1.
If not using sliding window LR context crop, config is as follows:
[1]test_cfg=dict(mode='whole')) If not using sliding window HR detail crop, config is as follows:
[2]hr_slide_inference=False,
In case 1, The number of forward passes is 160M?
*160M = 60M(Encoder(LR Crop)) + 20M(Decoder(LR Crop)) + 60M(Encoder(HR Crop)) + 20M(Decoder(HR crop))
Because if hr_slide_inference and sliding window inference are not used, the whole image is used for LR context crop and HR detail crop.
Also, am I wrong to think that HRDA is a resolution-based ensemble method?
@lhoyer
Appreciate your kindness. You're right. What I want to know is the forward pass.
Can you cross check if the calculation is correct?
Case1. If not using sliding window LR context crop, config is as follows:
[1]test_cfg=dict(mode='whole'))
If not using sliding window HR detail crop, config is as follows:[2]hr_slide_inference=False,
In case 1, The number of forward passes is 160M? *160M = 60M(Encoder(LR Crop)) + 20M(Decoder(LR Crop)) + 60M(Encoder(HR Crop)) + 20M(Decoder(HR crop))
Because if hr_slide_inference and sliding window inference are not used, the whole image is used for LR context crop and HR detail crop.
Also, am I wrong to think that HRDA is a resolution-based ensemble method?